Privacy-Cost Cooperative Games and Prefix-Sampling Data Valuation: An Adult Benchmark Study
Year: 2026
Type: article
Abstract: We formulate a cooperative data-sharing game with task utility, a stylized privacy cost and a coordination charge. Convexity and approximate core guarantees require explicit net-worth and error conditions. Ordered prefix sampling exposes individual marginals through public differences; the experiment evaluates a simulated interface. On one Adult split with 400 valuation and 600 evaluation records, a kNN recursion reportedly agrees with eight-record enumeration to 6.9e-16. At 128 permutations, simulated SecPSA has Spearman agreement 0.564 ± 0.036 with the recursion anchor, versus 0.572 ± 0.032 for Monte Carlo and 0.605 ± 0.017 for Data Banzhaf. Matched-permutation vectors reportedly agree exactly, with 6–16% simulated overhead. Privacy costs and institutional coalitions remain untested. Original utility definitions, code and seed-level outputs are required to audit the larger-pool anchor and statistical summaries.
Source: Systemic Analytics
DOI: 10.31181/sa44202698
Institutions: Bitlis Eren University, University of Cape Coast
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7234864520
